Looking back on LLMs

Hey everybody, and welcome to the 2,703 people joining from last week. A quiet one for me, amidst tears from watching Wales lose, again.

This is Ben's Bites - the AI newsletter that will tickle your circuits with its daily dose of AI puns and information.

Let's get to it.

Have a product, service, job, event, newsletter, app, book, movie, tool, or anything you'd like to share with over 26k subscribers? 

Advertise with us | Job board | Join Discord

🤌 Ben's Picks

1/

PhotoAI.io transforms your photos into a benchmark style

I’ve always fancied myself to be a hit in the modelling industry (😂 ofc not), and now I can basically turn myself into one with PhotoAI.io. This is the guy (Pieter Levels) that brought us those insanely popular personalised avatars last year. Now he's propelled that into photorealism - select a model photo and get your photos fully transformed into that style. Adjust your clothes, your stance, your setting, and more. Although, errr, it’s kind of catfisher heaven right!?

2/

ChatGPT is a blurry JPEG of the web

I’m a big fan of taking a step back and assessing reality when there’s a great deal of excitement frothing around something new (chatbot AI), and this article satisfied that hunger in me! They compare LLMs to a blurry JPEG image - something that’s unable to perfectly replicate existing information and can be wrong, but roughly represents the important bits of information that it received. With lots of talk around search engines shifting to chatbot styles, it’s important to ask ourselves the question: Do we actually want what essentially amounts to a lossy compression (ChatGPT), rather than something returning lossless results (Google search)?

For some more healthy scepticism, check out this one arguing that perhaps perfectly mimicking humans is the wrong approach to reach AGI; and this one stating that whatever we do, dark thoughts and expressions will always be lurking within chatbot technology, and we’re merely able to suppress these.

3/

Microsoft to integrate ChatGPT into Word, Powerpoint, and Outlook

Imeeeaaan, it’s kinda obvious that it’s gonna happen but still, it’s pretty fun to hear rumours this exciting hitting a platform as big as The Verge. They’re reporting that Microsoft has full intentions to integrate ChatGPT across all of its productivity services - Word, Powerpoint, Outlook, the lot. Imagine a world where word documents are continuously generating long-form suggestions, Powerpoint is able to design and generate unique visuals, and Outlook can respond to emails in your personal style. Wowzer.

🛠️ Cool Tools

• Contentinator - AI content generation directly in Figma. (link)

• Mokker - AI background replacement for instant professional product photos from a single product image. (link)

• Custom GPT - Build your own chatbot with your data. (link)

• Snape - Make job descriptions exciting with personalised and curated copy powered by AI. (link)

• Hire People - Send personalised LinkedIn invitations, instantly. (link)

• ExtractGPT - A browser extension for scraping data from structured & unstructured pages. (link)

• Sellesta - Make more money on Amazon with AI with optimization recommendations for your listing. (link)

• AIcyclopedia - The free AI encyclopaedia with 2000+ resources. (link)

• Cognify AI studio - Transform your photos into stunning designs with the power of AI. (link)

• ArxivGPT - Summarize an arXiv paper and provide key insights. (link)

• People AI - Boost your personal brand with your very own digital twin. (link)

• Autobackend - Create an AI-generated backend with just a few sentences. (link)

🤓 Miscellaneous

• COLLODI - Work in progress text-to-3D model using OpenAI's Point-E, Gepetto AI & Unreal Engine. (link)

• A quick and sobering guide to cloning yourself. (link)

• Imagining the QAnon of the AI era. (link)

• Microsoft to demo its new ChatGPT-like AI in Word, PowerPoint, and Outlook soon. (link)

• Google research, 2022 & beyond - Algorithmic advances. (link)

• ChatGPT frenzy sweeps China as firms scramble for home-grown options. (link)

• What if the future of AGI rhymes with the history of flight? (link)

• Inside the heart of ChatGPT’s darkness. (link)

• What would Joseph Schumpeter have made of Apple? (link)

• What if GPT had internal context on your business? (link)

• The coach behind Sam Altman, Naval, Brian Armstrong Shares the mistakes that all CEOs make. (link)

• Opera enters the generative AI space with new features in browsers and content apps. (official announcement - link, Verge coverage - link)

• The AI boom that could make Google and Microsoft even more powerful. (link)

• Reddit thinks AI chatbots will ‘complement’ human connection, not replace it. (link)

• What Google should really be worried about? (link)

• Some UX predictions for the coming age of LLMs. (link)

• Why I chose OpenAI over academia, part 2 - Reflections on the CS academic and industry job markets. (link)

• The next generation of large language models. (link, without paywall here)

• Searching the real images island - Understanding how diffusion-based generative models work. (link)

🎓 Learn

• Tutorial: Automating PPC client reports using Zapier and OpenAI. (link)

• Creating a user-flow diagram using ChatGPT that covers edge cases. (link)

• How to publish a children's book on a weekend using AI. (link)

• How to build a SwiftUI app that turns dog photos into classical paintings. (link)

🔬 Research

• The number of params needed to fine-tune Flan-T5-XXL is now 9.4M. About 7X fewer than AlexNet. (link)

• ControlNet - A neural network structure to control pre trained large diffusion models to support additional input conditions. (link)

• Code release for pix2pix-zero - Zero-shot image-to-image translation is out. (link)

• A novel binarized machine translation technique for transformers to address the problem of inflated dot-product variance. (link)

• Hindsight Instruction Relabeling (HIR) - Aligning language models with instructions without additional parameters or reward networks. (link)

• ViT-22B from Google - The currently largest vision transformer model at 22 billion parameters. (link)

• EVC from Microsoft - Towards real-time neural image compression with mask decay. (link)

• Language-aware multilingual machine translation with self-supervised learning from Meta. (link)

• Controllability-aware Skill Discovery (CSD) - Novel unsupervised skill discovery method which actively seeks complex, hard-to-control skills without supervision. (link)

• Project and Probe - Sample-efficient domain adaptation by interpolating orthogonal features. (link)

📰 Unclassifieds

• Covering the business of AI in a <5 min weekly read. New products to stay ahead and technical news translated. Your peak into the future.

• Tactical Arbitrage helps find items from 900+ websites to buy low and sell for a profit on Amazon using AI and image recognition. Use ER10 for a free 10 day trial.

• Zingtree, the no-code interactive decision tree platform, is hiring a Senior AI Engineer. (link)

Have a product, service, job, event, newsletter, app, book, movie, tool, or anything you'd like to share with over 26k subscribers? 

Advertise with us | Job board | Join Discord

🖼 AI images of the day

🤗 Share Ben's Bites

Send this to 1 AI-curious friend and receive my AI project tracker database!

or copy/paste this link: https://bensbites.beehiiv.com/subscribe?ref=PLACEHOLDER

⭐️ How did we do?

⭐️ REAL REVIEWS